The budget comes before the scene

A budget is five numbers written before the first asset exists: draw calls per frame, triangles on screen, texture memory in megabytes, JavaScript transferred in kilobytes, and the frame time you intend to hold. Without them, optimisation is an argument about taste in week nine, and every argument is won by whoever cares least about the phone.

A defensible starting point for a marketing-grade scene on mid-range mobile is something like 150 draw calls, 400,000 triangles, 60MB of texture memory after KTX2, and under 500KB of JavaScript for the 3D layer. Adjust them for your case, but adjust them deliberately and write down why.

Convert the frame rate into time and it gets concrete. 60fps is 16.7ms total, of which the browser wants a few for compositing, so your realistic ceiling is around 12ms, and a phone that is thermally throttled after two minutes gives you roughly half of what it gave you in the first ten seconds. Budget for the throttled device, because that is the one your visitor is holding.